Re: [PATCH v6 0/5] mm/sparse: Optimize memmap allocation during sparse_init()
From: Baoquan He
Date: Sat Jul 07 2018 - 22:09:26 EST
Hi Andrew,
Could you pick this series into mm tree so that it can catch 4.18ï
Thanks
Baoquan
On 06/28/18 at 02:28pm, Baoquan He wrote:
> This is v6 post.
>
> In sparse_init(), two temporary pointer arrays, usemap_map and map_map
> are allocated with the size of NR_MEM_SECTIONS. They are used to store
> each memory section's usemap and mem map if marked as present. In
> 5-level paging mode, this will cost 512M memory though they will be
> released at the end of sparse_init(). System with few memory, like
> kdump kernel which usually only has about 256M, will fail to boot
> because of allocation failure if CONFIG_X86_5LEVEL=y.
>
> In this patchset, optimize the memmap allocation code to only use
> usemap_map and map_map with the size of nr_present_sections. This
> makes kdump kernel boot up with normal crashkernel='' setting when
> CONFIG_X86_5LEVEL=y.
>
> The old version can be found below:
>
> v5:
> http://lkml.kernel.org/r/20180627013116.12411-1-bhe@xxxxxxxxxx
> v4:
> http://lkml.kernel.org/r/20180521101555.25610-1-bhe@xxxxxxxxxx
>
> v3:
> https://lkml.org/lkml/2018/2/27/928
>
> V1 can be found here:
> https://www.spinics.net/lists/linux-mm/msg144486.html
>
> Change log:
> v5->v6:
> Oscar found the redundant "struct mem_section *ms" definition and
> in the old patch 2/4, after deferring the clearing of section_mem_map.
> Clean them up in this version.
>
> Pavel pointed out that allocating memmap together for one node at
> one time should be a default behaviour for all ARCH-es. And if failed
> on large memory, it will drop to the fallback to allocate memmap
> for one section at one time, it shoult not break anything. Add
> patch 5/5 to remove CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER and clean
> up the related codes.
> v4->v5:
> Improve patch 3/4 log according to Dave's suggestion.
>
> Correct the wrong copy&paste of making 'nr_consumed_maps' to
> 'alloc_usemap_and_memmap' mistakenly which is pointed out by
> Dave in patch 4/4 code comment.
>
> Otherwise, no code change in this version.
> v3->v4:
> Improve according to Dave's three concerns which are in patch 0004:
>
> Rename variable 'idx_present' to 'nr_consumed_maps' which used to
> index the memmap and usemap of present sections.
>
> Add a check if 'nr_consumed_maps' goes beyond nr_present_sections.
>
> Add code comment above the final for_each_present_section_nr() to
> tell why 'nr_consumed_maps' need be increased in each iteration
> whether the 'ms->section_mem_map' need cleared or out.
>
> v2->v3:
> Change nr_present_sections as __initdata and add code comment
> according to Andrew's suggestion.
>
> Change the local variable 'i' as idx_present which loops over the
> present sections, and improve the code. These are suggested by
> Dave and Pankaj.
>
> Add a new patch 0003 which adds a new parameter 'data_unit_size'
> to function alloc_usemap_and_memmap() in which we will update 'data'
> to make it point at new position. However its type 'void *' can't give
> us needed info to do that. Need pass the unit size in. So change code
> in patch 0004 accordingly. This is a code bug fix found when tested
> the memory deployed on multiple nodes.
>
> v1-v2:
> Split out the nr_present_sections adding as a single patch for easier
> reviewing.
>
> Rewrite patch log according to Dave's suggestion.
>
> Fix code bug in patch 0002 reported by test robot.
>
> Baoquan He (5):
> mm/sparse: Add a static variable nr_present_sections
> mm/sparsemem: Defer the ms->section_mem_map clearing
> mm/sparse: Add a new parameter 'data_unit_size' for
> alloc_usemap_and_memmap
> mm/sparse: Optimize memmap allocation during sparse_init()
> mm/sparse: Remove CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
>
> mm/Kconfig | 4 --
> mm/sparse-vmemmap.c | 9 ++---
> mm/sparse.c | 109 ++++++++++++++++++++++++++++------------------------
> 3 files changed, 62 insertions(+), 60 deletions(-)
>
> --
> 2.13.6
>